Modeling and State Estimation of Autonomous Vehicles in Signalized Intersections

During automated driving in signalized intersections, decisions must be made by autonomous vehicles based on the system dynamics and the measurements provided by the sensors. The system dynamics is often affected by some uncertainty due to unmodeled parts of the system, the discrepancies in modeling the vehicle dynamics, and the adaptive timings of traffic lights according to traffic conditions. The measurements perceived by sensors about the traffic lights and the vehicle position are often noisy due to factors including various illumination conditions, incomplete shapes due to occlusion, very few pixels for detecting distant traffic lights, and motion blurring due to high-speed driving. These sources of uncertainty and noises can directly impact the vehicle's perception about the states of the traffic light and the vehicle, which can cause running the red light or rear-end crashes from the sudden brakes. In this paper, we propose a hidden Markov model (HMM) representation of the dynamical system that takes into account all sources of uncertainty and noisy measurements. The proposed framework estimates the state of vehicle and traffic light via sequential noisy observations through a particle filtering scheme. The reliability of the proposed framework is demonstrated in the numerical experiments.